January 3rd, 2006

Help needed on Sun POD design and cost

Posted by Paul Murphy @ 2:06 am

Categories: Enterprise Policy, General, Hardware, Sun Pod

Tags:

I think 2006 will go down in history as a year of phenomenal change in IT: more dramatic and more through going than anything we’ve seen before. By the end of the year we’ll probably have seen the end of Apple’s Intel affair, IBM will have publicly committed to moving all the hardware they sell to Linux on Cell, Microsoft will have developer pre-release information out on a networked Windows successor that only runs on Xenon, and Sun, of course, is already in the field with their "cool threads" throughput computing initiative.

One of the most interesting questions affecting What I need you to help me with is finding ways to make it better, or cheaper. all this revolves on how quickly users of existing technologies will adopt the new hardware and OS tools and thereby drive development of the software base needed for each vendor to succeed.

Since I use Sun gear and they’re first in the field, I want to start a series on porting related issues by creating a standardized "Sun Pod" - consisting of Sun CMT hardware and Solaris software - for use as the reference, or target, environment.

The basic idea is to create what Sun doesn’t offer: a configurable, deskside, package incorporating the computer, storage, power, and networking.

Here’s what’s needed on the hardware side:

1. a lockable rack mount with internal power and network cabling to hold everything neatly together;

    

2. a sufficiently large bunch of disks for use with ZFS and zRAID;

    

3. a collection of traditional I/O devices giving R/W access to, at least: CDs, DVDs, and DAT style tapes

    

4. at least one T1 series ("Niagara") computer

    

5. network isolation and connectivity

    

6. a UPS capable of filtering power during normal operations and giving at least three hours of uptime during external power interruptions;

Since we’ll be applying this pod to a fairly wide range of applications for small and medium sized organizations we need it to be highly configurable. At the same time, however, we want it to be a bit of a masterpiece - like the fabulous the one hoss shay described by Oliver Wendell Holmes:

Now in building of chaises, I tell you what,
There is always somewhere a weakest spot, –
In hub, tire, felloe, in spring or thill,
In panel, or crossbar, or floor, or sill,
In screw, bolt, thoroughbrace, — lurking still,
Find it somewhere you must and will, –
Above or below, or within or without, –
And that’s the reason, beyond a doubt,
A chaise breaks down, but does n’t wear out.

But the Deacon swore (as Deacons do,
With an "I dew vum," or an "I tell yeou")
He would build one shay to beat the taown
‘N’ the keounty ‘n’ all the kentry raoun’;
It should be so built that it could n’ break daown:
"Fur," said the Deacon, "’t ’s mighty plain
Thut the weakes’ place mus’ stan’ the strain;
‘N’ the way t’ fix it, uz I maintain,
Is only jest
T’ make that place uz strong uz the rest."

The T1 computers have a long list of high end built-in RAS features including:

• ECC throughout (including L2 cache, and the integer and floating point registers)

   

• Parity/Retry logic on Data TLB, instruction TLB, and both data and instructions sides of L1 Cache

   

• Chipkill, Address parity, Memory Scrubber (including L2 cache) and DRAM sparing

Matching the rest of the pod to that requires that the JBOD be hot serviceable, the network gear have extensive redundancy, the rack power and cabling distribution system has to be of high quality, the UPS has to have first class power filtration, and be easily connectible to an external power generator.

In addition, because emergencies like building fires and regional network outages do occur, we will run into situations in which it’s imperative that operations on one pod can be seamlessly continued on another situated in a physically different building, city, or state.

According to Holmes "the Deacon inquired of the village folk / Where he could find the strongest oak" and, guess what? if this is the global village: you’re folk. Here’s a very rough POD configuration done using Sun components where possible, and what I need you to help me with is finding ways to make it better, or cheaper, or both without giving up reliability or flexibility.

 

1. Sun Fire T2000 Server ($27,160)

   1 1.0-GHz 8 Core UltraSPARC T1 Processor
   3-MB Level-2 Cache
   16-GB Memory
   4 73-GB 10000 RPM SAS Disk Drives
   1 DVD-ROM/CD-RW Drive
   4 10/100/1000 Ethernet Ports
   1 Serial Port
   3 PCI Express Slots
   2 PCI-X Slots
   2 PCI UltraSCSI 320 Controllers 2 Power Supply Units
   Java Enterprise System Software Pre-Installed
   Solaris 10 Operating System Pre-Installed
   Three years "platinum" (2 hour on site) support.

    

2. Sun Ray 170, 17 inch LCD, ($699)

    

3. Sun 900-36N Rack ($7,538)

   NEBS Earthquake Certified
   External Dimensions: 23.5 in. x 35.4 in. x 74 in.
   Rack Unit Dimensions 19 in. x 900mm x 1.75 in.
   Rack Units 36
   Lockable Rear Door
   Lockable Front Door
   Power System
   Cable Management Bracket
   Stabilizer Anti-Tilt Bar/Floor Brackets

    

4. Two Sun D240 Media Trays ($17,228)

   Six 146GB US3 15K RPM drives
   Two 72GB 3MB/Sec 4MM DAT drives
   Two Rackmount kits

    

5. Missing: router/switch and UPS

Since this pod is intended for use as a general target environment, I’ve made it a middle of the road, run-time only, kind of system. Before getting to the missing pieces consider these additional configuration options:

1. for use business uses beyond the plug and go operation of well defined applications: add a 24" Sun Ray 1g ($2,699) and a stripped, four core, 8GB, T2000 with dual 73GB disks for $7995.

    

2. for use in larger production environments upgrade the T2000 to 1.2Ghz with 32Gb of RAM ($9,000); add a second one at $34,891 inclusive of a FC controller set and three years of two hour on site support; and add a dual port Sun 3510 1.4TB FC array for a list price of $32,720.

In the basic system configuration above I’ve used a rack so absurdly expensive the designers probably thought of it as a T-2000 Exoskeleton (to match the T-800/850 Endoskeletons), and it’s about twice as big as we need too. So, first question: bearing mind that we need a minimum of 34 inches of depth and about 18U, can anyone suggest some better choices?

I’ve omitted the power and network gear from the configuration, simply because I don’t know what makes sense.

The UPS requirements are pretty simple. The basic T2000 draws no more than 400 watts, and the two media trays, fully loaded get about that too - i.e. about 1,200 watts total before networking gear. Lets assume we want to be able to double this before upgrading the UPS and need continuous filtration plus three hours of run-time with hookups in place for an external power generator. Suggestions?

The networking requirements are harder. At a minimum, we need to interface the pod both to the web and to the typical small to medium size business network - figure from 500 to 2,500 desktops in subnets of 50 to 100, with at least some of them not local. At maximum - well, we’d like to be able to consider things like dial-up ADSL operation and telecom integration. Suggestions?

Please bear in mind that we’ll also want to be able to configure two pods, one in each of two widely separated locations, to provide redundancy and failover capabilities ranging from simple database replication through heartbeat clustering right up to the level claimed for the Kabira continuous availability suite.

Notice, finally, that the DVD R/W combined with the two DAT drives do not amount to a serious off-site backup capability. Their purpose is to support data transfer, allow off-line (but on site) storage for some user files, and provide the ability to produce daily or weekly compressed data files intended more to meet contractual obligations on off site data storage than to provide a practical recovery option.

One note: if you’re a vendor, please don’t pollute talkback with long sales pitches - just the facts ok? remember you can reference your own site and PDFs in the paragraph or two you place in talkbacks.